Warm spring reduced carbon cycle impact of the 2012US summer droughtSebastian Wolfa,b,1, Trevor F. Keenanc,2, Joshua B. Fisherd, Dennis D. Baldocchia, Ankur R. Desaie,Andrew D. Richardsonf, Russell L. Scottg, Beverly E. Lawh, Marcy E. Litvaki, Nathaniel A. Brunsellj,Wouter Petersk,l, and Ingrid T. van der Laan-Luijkxk

aDepartment of Environmental Science, Policy, and Management, University of California, Berkeley, CA 94720; bDepartment of Environmental SystemsScience, ETH Zurich, 8092 Zurich, Switzerland; cDepartment of Biological Sciences, Macquarie University, Sydney, NSW 2109, Australia; dJet PropulsionLaboratory, California Institute of Technology, Pasadena, CA 91109; eAtmospheric and Oceanic Sciences, University of Wisconsin–Madison, Madison, WI53706; fDepartment of Organismic and Evolutionary Biology, Harvard University, Cambridge, MA 02138; gUS Department of Agriculture, AgriculturalResearch Service, Southwest Watershed Research Center, Tucson, AZ 85719; hDepartment of Forest Ecosystems and Society, Oregon State University,Corvallis, OR 97331; iDepartment of Biology, University of New Mexico, Albuquerque, NM 87131; jDepartment of Geography, University of Kansas,Lawrence, KS 66045; kDepartment of Meteorology and Air Quality, Wageningen University, 6708 PB Wageningen, The Netherlands; and lCentre for IsotopeResearch, University of Groningen, 9747 AG Groningen, The Netherlands

Edited by Susan E. Trumbore, Max Planck Institute for Biogeochemistry, Jena, Germany, and approved March 22, 2016 (received for review October 3, 2015)

The global terrestrial carbon sink offsets one-third of the world’sfossil fuel emissions, but the strength of this sink is highly sensitiveto large-scale extreme events. In 2012, the contiguous United Statesexperienced exceptionally warm temperatures and the most severedrought since the Dust Bowl era of the 1930s, resulting in substantialeconomic damage. It is crucial to understand the dynamics of suchevents because warmer temperatures and a higher prevalence ofdrought are projected in a changing climate. Here, we combine anextensive network of direct ecosystem flux measurements withsatellite remote sensing and atmospheric inverse modeling to quantifythe impact of the warmer spring and summer drought on biosphere-atmosphere carbon and water exchange in 2012. We consistently findthat earlier vegetation activity increased spring carbon uptake andcompensated for the reduced uptake during the summer drought,which mitigated the impact on net annual carbon uptake. The earlyphenological development in the Eastern Temperate Forests played amajor role for the continental-scale carbon balance in 2012. The warmspring also depleted soil water resources earlier, and thus exacerbatedwater limitations during summer. Our results show that the detrimen-tal effects of severe summer drought on ecosystem carbon storagecan be mitigated by warming-induced increases in spring carbon up-take. However, the results also suggest that the positive carbon cycleeffect of warm spring enhances water limitations and can increasesummer heating through biosphere–atmosphere feedbacks.

seasonal climate anomalies | carbon uptake | ecosystem fluxes |biosphere–atmosphere feedbacks | eddy covariance

An increase in the intensity and duration of drought (1, 2),along with warmer temperatures, is projected for the 21stcentury (3). Warmer and drier summers can substantially reducephotosynthetic activity and net carbon uptake (4). In contrast,warmer temperatures during spring and autumn prolong theperiod of vegetation activity and increase net carbon uptake intemperate ecosystems (5), sometimes even during spring drought(6). Atmospheric CO2 concentrations suggest that warm-spring–induced increases in carbon uptake could be cancelled out by theeffects of warmer and drier summers (7). However, the extentand variability of potential compensation on net annual uptakeusing direct observations of ecosystem carbon exchange have notyet been examined for specific climate anomalies.In addition to perturbations of the carbon cycle, warmer spring

temperatures can have an impact on the water cycle by increasingevaporation from the soil and plant transpiration (8–10), whichreduces soil moisture. Satellite observations suggest that warmerspring and longer nonfrozen periods enhance summer drying viahydrological shifts in soil moisture status (11). Climate model sim-ulations also indicate a soil moisture–temperature feedback betweenearly vegetation green-up in spring and extreme temperatures in

summer (12, 13). Soil water deficits during drought impose areduction in stomatal conductance, thereby reducing evapora-tive cooling and thus increasing near-surface temperatures (14).Stomatal closure also has a positive (enhancing) feedback with at-mospheric water demand by increasing the vapor pressure deficit(VPD) of the atmosphere (15). The vegetation response thus playsa crucial role for temperature feedbacks during drought (16).Given the opposing effects of concurrent warmer spring and sum-

mer drought, and an increased frequency of these anomalies projecteduntil the end of this century (SI Appendix, Fig. S1), it is imperativeto understand (i) the response of the terrestrial carbon balance and(ii) the interaction of carbon uptake with water and energy fluxes thatare associated with these seasonal climate anomalies.The year 2012 was among the warmest on record for the

contiguous United States (CONUS), which experienced one of

Significance

Carbon uptake by terrestrial ecosystems mitigates the impactof anthropogenic fossil fuel emissions on atmospheric CO2concentrations, but the strength of this carbon sink is highlysensitive to large-scale extreme climate events. In 2012, theUnited States experienced the most severe drought since theDust Bowl period, along with the warmest spring on record.Here, we quantify the impact of this climate anomaly on thecarbon cycle. Our results show that warming-induced earliervegetation activity increased spring carbon uptake, and thuscompensated for reduced carbon uptake during the summerdrought in 2012. This compensation, however, came at the costof soil moisture depletion from increased spring evapotrans-piration that likely enhanced summer heating through land-atmosphere coupling.

Author contributions: S.W., T.F.K., J.B.F., D.D.B., A.R.D., and A.D.R. designed research; S.W.performed research; S.W. analyzed data; S.W., T.F.K., J.B.F., D.D.B., A.R.D., A.D.R., R.L.S., B.E.L.,M.E.L., N.A.B., W.P., and I.T.v.d.L.-L. wrote the paper; S.W. compiled the flux datasets; T.F.K.performed the flux data gap-filling and partitioning; J.B.F. compiled and processed the Mod-erate Resolution Imaging Spectroradiometer (MODIS) and Coupled Model IntercomparisonProject Phase 5 (CMIP5) data; A.R.D. performed the CarbonTracker analysis (CT2013B); andW.P.and I.T.v.d.L.-L. performed the CarbonTracker analysis (CTE2014 and CTE2015).

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The eddy-covariance data are available in the AmeriFlux data archive atthe Carbon Dioxide Information Analysis Center at the Oak Ridge National Laboratory(cdiac.ornl.gov/ftp/ameriflux/data).

See Commentary on page 5768.1To whom correspondence should be addressed. Email: sewolf@ethz.ch.2Present address: Earth and Environmental Sciences, Lawrence Berkeley National Labora-tory, Berkeley, CA 94720.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental.

5880–5885 | PNAS | May 24, 2016 | vol. 113 | no. 21 www.pnas.org/cgi/doi/10.1073/pnas.1519620113

Dow

nloa

ded

by g

uest

on

June

14,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1519620113&domain=pdfhttp://cdiac.ornl.gov/ftp/ameriflux/datamailto:sewolf@ethz.chhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplementalwww.pnas.org/cgi/doi/10.1073/pnas.1519620113

the most severe droughts since the Dust Bowl era of the 1930s(17, 18). The drought caused substantial economic damage,particularly for agricultural production (SI Appendix). Annual meantemperatures were 1.8 °C above average, with the warmest spring(+2.9 °C) and second warmest summer (+1.4 °C) in the period of1895–2012 (19). Precipitation deficits started to evolve in Mayacross the Great Plains and the Midwest (17), but eventually af-fected more than half of the United States (20). By July, 62% of theUnited States experienced moderate to exceptional drought, whichwas the largest spatial extent of drought for the United States sincethe Dust Bowl era (19). Severe drought conditions with depletedsoil moisture persisted throughout summer, and unprecedentedprecipitation deficits of 47% below normal for May throughAugust were observed in the central Great Plains (17).Here, we analyze the response of land-atmosphere carbon and

water exchange for major ecosystems in the United States duringthe concurrent warmer spring and summer drought of 2012 at theecosystem, regional, and continental scales. We combine directmeasurements of land-atmosphere CO2, water vapor, and energyfluxes from 22 eddy-covariance (EC) towers across the UnitedStates (SI Appendix, Fig. S2 and Table S1) with large-scale satelliteremote-sensing observations of gross primary production (GPP),evapotranspiration (ET), and enhanced vegetation index (EVI)derived from the space-borne Moderate Resolution ImagingSpectroradiometer (MODIS), and estimates of net ecosystem pro-duction (NEP; i.e., net carbon uptake) from an atmospheric CO2inversion (CarbonTracker, CTE2014). This comprehensive suite ofstandardized analyses across sites and data streams was crucial toconstrain the impact of such a large-scale drought event with bot-tom-up and top-down approaches (21), and something only a fewsynthesis studies have achieved so far (4, 22).We test the hypothesis that increased carbon uptake due to warm

spring offset the negative impacts of severe summer drought during2012, and examine the relationship between early-spring–inducedsoil water depletion and increased summer temperatures. Whenusing the term “drought,” we refer to precipitation deficits thatresulted in soil moisture deficiencies (9).

ResultsEvidence from in Situ Measurements. At the site scale, spring(March–May) 2012 anomalies of net carbon uptake significantlyincreased with temperature (Fig. 1A), on average by 15± 13 g Cm−2

season−1 (mean ± uncertainty: +29%, n = 22; SI Appendix, TableS2) across all sites, relative to the baseline 2008–2010. This in-crease was linked to earlier and increased vegetation activity(SI Appendix, Table S3). Some sites also experienced pre-cipitation deficits during spring, but the effect on net carbonuptake across sites was not significant (SI Appendix, Fig. S3). Incontrast, summer (June–August) net carbon uptake decreased atmost sites by 24 ± 18 g C m−2 season−1 (−17%, n = 22), and these

reductions were highly correlated with drought intensity (R2 = 0.64,P < 0.001; Fig. 1B).Of all sites, 13 (59%) experienced summer drought conditions

with precipitation deficits of at least 10% or 19–326 mm season−1(SI Appendix, Fig. S2 and Table S4). Subsequently, these sites wereused to investigate the process level impact of concurrent warmerspring and summer drought at the ecosystem scale.Spring net carbon uptake increased on average by 25 ±

11 g C m−2 season−1 (+103%, n = 13), and summer uptake de-creased by 32 ± 18 g C m−2 season−1 (−25%) at sites that experi-enced summer drought (Fig. 2). Consequently, a warmer springcompensated, on average, for 78% of drought-related reductions insummer net carbon uptake, and reduced the impact of summerdrought on annual net carbon uptake to −17 ± 18 g C m−2 season−1(−11% compared with baseline; Fig. 2). A consistent, althoughsmaller, partial compensation pattern was detected for GPP atsummer drought-affected sites from both direct EC measure-ments (40% compensation) and satellite-derived MODIS GPPestimates (39% compensation; SI Appendix, Figs. S4 and S5).

Large-Scale Carbon Cycle Impact. At the regional scale, the GreatPlains ecoregion (23) experienced the strongest drought condi-tions during summer (SI Appendix, Table S5), with reductions of123 g C m− 2 season−1 (−36%) in GPP from MODIS estimates(Fig. 3 and SI Appendix, Table S6). NEP estimates from inversemodeling by CarbonTracker (CTE2014) showed reductions of71 g Cm−2 season−1 (−72%) in net carbon uptake in the Great Plains,which exceeded the reductions in other ecoregions (SI Appendix,Table S6). Substantial summer reductions were also found forthe Northern Forests ecoregion (GPP: −114 g C m−2 season−1,NEP: −38 g C m−2 season−1) and for the Eastern Temperate Forestsecoregion (GPP:−86 g Cm−2 season−1, NEP:−17 g Cm−2 season−1),which includes large parts of the Midwest. During the warm springof 2012, these three ecoregions also had among the strongest in-creases in GPP, ranging from 32 to 75 g Cm−2 season−1 (SI Appendix,Table S6), which was particularly evident across the Appalachians(Fig. 3A) because tree phenology benefitted substantially fromwarmer temperatures during spring (SI Appendix, Fig. S6). TheEastern Temperate Forests showed by far the highest increase inspring NEP (69 g C m−2 season−1). Annual net carbon uptake in theGreat Plains was reduced by 46 g C m−2 y−1, making the ecoregionclose to carbon neutral in 2012 (−5 g C m−2 y−1 vs. baseline: 41 ±24 g C m−2 y−1). In contrast, the Eastern Temperate Forests hadthe highest net uptake during the last decade in 2012 (136 g C m−2

y−1) and were outside the interannual variability from 2001 to2011 (58 ± 19 g C m−2 y−1).Across the entire CONUS, NEP estimates by CarbonTracker

showed total increases of 0.24 Pg C season−1 during spring andreductions of 0.23 Pg C season−1 in summer 2012 (see SI Appendix,Table S6 for mean values in g C m−2). In 2012, spring net carbonuptake (0.76 Pg C season−1) was the highest and summer carbon

A BFig. 1. Seasonal climate anomalies increased netcarbon uptake in spring but led to reductions insummer throughout the United States in 2012. Fluxtower-derived seasonal anomalies of NEP (g C m−2)during spring (A) and summer (B; related to droughtstress via the ESI) in 2012 relative to the baselineof 2008–2010. Symbols and colors denote the In-ternational Geosphere-Biosphere Program (IGBP)land-use classes, which are provided with the sitenames and abbreviations in SI Appendix, Table S1.Error bars denote the uncertainties in the fluxanomalies. Dashed lines represent the confidenceinterval of the ordinary least squares mean regression(bold line). The summer anomaly at the site KON isout of scale (SI Appendix, Table S10) and was omit-ted from display only. Further details are provided inSI Appendix, Figs. S3, S15, and S16).

Wolf et al. PNAS | May 24, 2016 | vol. 113 | no. 21 | 5881

ENVIRONMEN

TAL

SCIENCE

SSU

STAINABILITY

SCIENCE

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

June

14,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdf

uptake was the lowest (0.38 Pg C season−1), and both seasons wereclearly outside the interannual variability across the CONUSduring the last decade (SI Appendix, Table S7). Due to the com-pensation of drought-induced reductions in summer by increasesfrom warm spring, and lower carbon losses during winter and fall(Fig. 3H), net annual carbon uptake across the CONUS was aboveaverage in 2012 (0.11 Pg C y−1). Net annual carbon uptake in 2012(0.33 Pg C y−1) was the second highest estimated by Carbon-Tracker since 2001 and outside the interannual variability of thebaseline (0.22 ± 0.05 Pg C y−1; SI Appendix, Table S7).

Water and Energy Flux Feedback. Across all EC sites (n = 22),summer increases in sensible heat (H) flux were highly correlatedwith reductions in ET or latent energy (LE; R2 = 0.54, P < 0.001;SI Appendix, Fig. S7). At sites with summer drought, we observedLE increases of 38 ± 15 MJ m−2 season−1 (mean ± uncertainty:+14%, n = 13) during early spring and reductions of −66 ±24 MJ m−2 season−1 (−12%) during summer 2012 relative tobaseline (SI Appendix, Fig. S8). This reduction in evaporativecooling was associated with soil water limitations and elevated VPD(SI Appendix, Table S8) causing stomatal closure. Increased ETduring spring and reduced ET during summer were also evidentfrom MODIS, both for these sites and at larger regional to conti-nental scales (SI Appendix, Figs. S9 and S6). At the sites with sum-mer drought, our results indicate that the warming-induced increasein vegetation activity during spring caused an earlier depletion ofwater resources (SI Appendix, Figs. S8 and S9), and thus contributedto soil water limitations during summer. This depletion resulted in arelative heating contribution from increased H during the period ofpeak insolation in summer (SI Appendix, Fig. S8).To evaluate the magnitude of a potential summer heating

enhancement by earlier spring vegetation activity, we quantifiedthe impact for sites located in the center of the drought in theMidwest (Fig. 4). These sites had summer precipitation deficitsof −300 ± 26 mm season−1 (−74%, mean ± SD), and springtemperatures were warmer by 4.8 ± 1.9 °C (SI Appendix, Fig.

S10). Earlier vegetation activity was evident in site observationsof GPP (Fig. 4C) and satellite EVI (SI Appendix, Fig. S11),resulting in increased LE, and induced an earlier drawdown insoil moisture (Fig. 4D). Daily soil moisture was highly and neg-atively correlated to LE (R2 = 0.72, P < 0.001) and GPP (R2 =0.73, P < 0.001) during spring 2012. At the beginning of summer,cumulative ET since January 1 was, on average, 40 mm (98MJ m−2,26%) higher, which was a 12-d forward shift in the total amountof evaporated water compared with the baseline. During sum-mer, evaporative cooling through LE was reduced by −201 ±32 MJ m−2 season−1 (−24%), with the majority of excess energyreleased through increased H (178 ± 19 MJ m−2 season−1 or107%; Fig. 4A). This increase in H had a relative heating effect(Fig. 4B) that exceeded elevated incoming shortwave radiation(SI Appendix, Fig. S12) and likely contributed to anomaloussurface heating of 2.0 ± 1.7 °C (SI Appendix).

DiscussionOur results consistently show increased carbon uptake in springand substantial reductions during summer in 2012 across in-dependent observations at the site scale and at the regional tocontinental scale. This study provides further evidence from di-rect observations for an offset in summer reductions by a warmerspring (7). Increased vegetation activity during the warm springin 2012 (24) compensated for reductions in net carbon uptake bythe severe summer drought. Consequently, the CONUS arearemained a carbon sink of 0.33 Pg C y−1 during 2012. In com-parison, the European 2003 summer drought resulted in a car-bon source of 0.50 Pg C y−1, equivalent to 4 y of carbon uptakeacross Europe (4). A longer lasting drought in western NorthAmerica reduced carbon uptake by 0.03–0.30 Pg C y−1 during2000–2004 but did not turn this region into a carbon source (22).More recently, the 2011 Texas drought was reported to havereduced net carbon uptake by 0.23 Pg C y−1, which was thelargest anomaly in this region since 1950 (25).Regional estimates of net carbon uptake in 2012 showed that

71% of the total summer reductions across the CONUS origi-nated from the Great Plains (−0.16 Pg C season−1), whereas74% of the increased spring uptake originated from the EasternTemperate Forests (0.18 Pg C y−1). It can therefore be arguedthat the higher spring uptake in the Eastern Temperate Forestswas the main reason that the large-scale 2012 summer droughtdid not result in reductions of net annual carbon uptake acrossthe CONUS. This spring compensation emphasizes the impor-tant ecosystem service of forests for climate and their contrastingresponse to climate anomalies compared with grasslands (6, 16).Ensemble EC site-scale measurements showed good agreementwith the continental-scale estimates by MODIS and Carbon-Tracker, suggesting that this ensemble of sites represents the large-scale impact of the 2012 climate anomaly across the CONUS quitewell. Our results show that the carbon cycle anomaly during springand summer 2012 was outside the range of data uncertainty (for ECsites; Fig. 2A) and also outside the large-scale interannual variability(for MODIS and CarbonTracker CTE2014) compared with thebaseline (Fig. 3 G and H). Although EC measurements across allsites were not available before 2008, long-term observations fromMODIS and CarbonTracker provide evidence that the baselineyears were within the 2001–2011 seasonal variability for carbonfluxes across the CONUS, yet higher during summer (SI Appendix,Table S7). Long-term precipitation data show that summer wasslightly wetter (+3%) during the baseline period, although withsimilar variability compared with the reference period (SI Appendix,Fig. S13 and Table S5). Consequently, by comparing the extremelydry summer of 2012 with a wetter than normal baseline period, wepotentially overestimate the carbon cycle impact of the 2012 event.Using MODIS and CarbonTracker data since 2001, we estimatethat the potential bias from our baseline selection for the conti-nental-scale impact of 2012 is 11% or less seasonally and 10% orless annually (SI Appendix, Table S7). Given the magnitude of the2012 drought, uncertainties due to baseline selection thus play aminor role for the large-scale impact assessment.

−10

0

10

20

30

40

50

Net

Car

bon

Upt

ake

(NE

P)

Spring Summer FallΔT: 2.4 °C 1.1 °C 0.7 °C

Annual1.5 °C

Baseline2012

A

−20−10

01020

ΔNE

P

more uptake

less uptake

25±11 g C −32±18 g C −11±11 g C −17±18

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec−40−20

02040

ΔP

wet

dry

2% −43% −24% −29%C

B

Fig. 2. Net carbon uptake of concurrent warm spring and summer droughtin 2012. (A) Ensemble mean of EC-measured monthly NEP (g C m−2 mo−1) for2012 (red) and baseline (black) at sites that experienced drought duringsummer 2012 (n = 13). Numbers atop show the mean seasonal temperature(T) anomalies in 2012 relative to the baseline of 2008–2010. (B) Anomalies ofNEP (g C m−2 mo−1) in 2012 relative to the mean baseline; numbers atopdenote the seasonal anomalies (g C m−2) and their uncertainties, which werederived from Monte-Carlo simulations of monthly fluxes (also shading in A).(C) Anomalies of monthly precipitation (mm mo−1) in 2012 relative to base-line; numbers atop show seasonal anomalies. Similar ensemble analyses forGPP and SPI can be found in SI Appendix, Fig. S4.

5882 | www.pnas.org/cgi/doi/10.1073/pnas.1519620113 Wolf et al.

Dow

nloa

ded

by g

uest

on

June

14,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1519620113

Due to the coupling of ecosystem carbon and water fluxes, thewarmer spring also increased ET, and thus depleted soil waterresources earlier. In 2012, soil water recharge from precipitationdid not occur (Fig. 4D); thus, the increased ET in spring furthercontributed to drought-related soil water limitations (primarily fromreduced precipitation) during summer. Evaporative cooling (fromLE) inhibits H flux from ecosystems to the atmosphere and preventsextreme heating on the surface and in the lower atmosphere duringthe period of peak insolation in summer (baseline shown in Fig.4A). The reduction in ET due to water limitations during summer2012 weakened this cooling effect, shifting the partitioning ofavailable energy toward H, and thus contributed to a positive (en-hancing) ecosystem heating feedback (Fig. 4B). Consequently, thepositive effect of higher carbon uptake during the warmer springcomes at the cost of soils drying out earlier, and thus potentiallyexacerbating drought and increasing summer heating in regions withlimited soil moisture reserves.Drier soils cause feedback mechanisms that amplify the

coupling between drought and heat when ET is inhibited (26).Previous evidence for such heating feedback amplification fromspring soil-moisture deficits was shown for European summerheat waves and also indicated that early summer precipitation

can mitigate the effect (13). The central Great Plains, whichincludes parts of the Midwest, is a region with strong biosphere-atmosphere coupling during summer (i.e., soil moisture also hassubstantial impacts on local precipitation) (27). Soil moisturedeficits thus have a positive (enhancing) feedback on droughtconditions in this region, highlighting the implications of po-tential water depletion from earlier vegetation activity duringwarm spring. Although our analyses cannot quantify the actualheating contribution from the ecosystem feedback during the2012 summer drought, these results provide further evidence fora linkage between earlier vegetation activity during a warmerspring, the associated depletion of soil water, and the enhancedheating from reductions in evaporative cooling (28) during sub-sequent summer drought (29). The combination of warm springand summer drought can also affect species composition basedon plant ecophysiology (SI Appendix).Although exceptional for the past century, it was found that

the sequence of atmospheric circulation patterns that caused the2012 drought resulted mostly from natural variability in climate(17). However, large-scale droughts, such as the 2012 drought,are projected to become more prevalent with global warming (18,20). Climate projections from the Coupled Model Intercomparison

A Spring B Summer C Annual G

PP (M

OD

IS)

D

NEP

(CTE

2014

) E F

G

Seas

onal

Cyc

le

H

0505−2012 Anomaly (g C m 2 mo 1)

0.0

0.2

0.4

0.6

0.8

1.0

1.2

Gro

ss P

rimar

y P

rodu

ctio

n (G

PP

) Spring Summer Fall Annual2001−2011Baseline2012

CONUS.MODIS

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

−0.2

0.0

0.2

GP

P

more uptake

less uptake

0.27 Pg C −0.59 Pg C −0.17 Pg C −0.38

−0.2

0.0

0.2

0.4

Net

Car

bon

Upt

ake

(NE

P)

Spring Summer Fall Annual2001−2011Baseline20122012 unc. est.

CONUS.CTE2014

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

−0.2

0.0

0.2

NE

P

more uptake

less uptake

0.24 Pg C −0.23 Pg C 0.06 Pg C 0.11

Fig. 3. Increased carbon uptake in the eastern United States during the warm spring of 2012 compensated for large reductions by the summer drought in theMidwest. Spatiotemporal anomalies are shown for GPP (g C m−2 mo−1) from MODIS (A–C) and NEP (g C m−2 mo−1) from CarbonTracker (CTE2014; D–F) duringspring, summer, and annually in 2012 relative to the baseline of 2008–2010. Red/orange colors indicate negative anomalies (reductions), and green/blue colors showpositive anomalies (increases). The seasonal cycle of total monthly GPP (G) and NEP (H) (both in Pg C) for the long-term mean (2001–2011, black dashed line), thebaseline (black line), and 2012 (red line) integrated across the CONUS. The red dotted line for NEP indicates an uncertainty estimate of CTE2014, based on the in-dependent model run CTE2015. The red shading indicates the 2012 uncertainty range between both model runs. The gray shading indicates the interannual variability(SD) during the baseline. Colored shading in lower panels shows the 2012 anomalies (Pg C) relative to the mean baseline; numbers atop denote seasonal anomalies.Further maps with spatial drought patterns (ESI, SPI), vegetation activity (EVI), and ET can be found in SI Appendix, Fig. S6.

Wolf et al. PNAS | May 24, 2016 | vol. 113 | no. 21 | 5883

ENVIRONMEN

TAL

SCIENCE

SSU

STAINABILITY

SCIENCE

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

June

14,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdf

Project Phase 5 (CMIP5) show warmer spring and drier summermean conditions by the end of this century across the CONUS(SI Appendix, Fig. S1), implying that such years will likely occurmore frequently. Whether the response observed in 2012 is repre-sentative of what can be expected under future climate changeremains subject to large uncertainty. For instance, CO2-inducedincreases in plant water-use efficiency are expected to affect plantfunction strongly, and could influence the effects of drought (30).Because the uncertainties in climate projections are linked to un-certainties in carbon cycle feedbacks (31), it is important to betterunderstand the response of the biosphere to seasonal climateanomalies and their effects on the annual carbon balance. Al-though the impact of a concurrent warmer spring and summerdrought depends on the specific climate anomalies (SI Appendix),the 2012 event serves as an example that enables the quantificationof the impact of such climate anomalies on the carbon and watercycles, and the surface energy budget.

ConclusionsWe conclude that the warm spring reduced the impact of the 2012summer drought on net annual carbon uptake across theUnited States. Regional differences played a major role for thecontinental-scale carbon balance in 2012: increased springuptake in the Eastern Temperate Forests largely compensatedfor reduced summer uptake in the Great Plains, emphasizing,in part, the ecosystem service value of these forests. However,the positive carbon cycle effect of a warmer spring and earliervegetation activity also had potentially detrimental side effectson the water cycle by increasing ET. Consistent with previousevidence on the water cycle effects of spring warming (11), ourresults suggest that the earlier depletion of soil water can in-crease ecosystem vulnerability to drought and exacerbatewarmer land surface temperatures through reduced evaporativecooling (28) during summer. Further research is needed to

understand such vegetation–climate feedbacks better, becausethey can have large implications for the management of waterresources and can potentially further increase the risk of watershortages during the period of highest demand in midsummer(8). Both the magnitude of the spring compensation effect forthe carbon cycle and hydrological feedbacks of earlier vegetationactivity highlight the importance of accurately predicting eco-system responses to spring warming.

Materials and MethodsWe analyzed the impact of concurrent warmer spring and summer droughton land-atmosphere carbon and water exchange across the United States in2012 using (i) direct measurements of EC, (ii) satellite remote-sensing ob-servations from MODIS, and (iii) estimates from the atmospheric inversemodel CarbonTracker.

Baseline Selection. Anomalies of 2012 were derived relative to the baseline of2008–2010 (as anomaly = 2012 − baseline) for a consistent reference periodacross all data streams. The selection of this baseline was constrained by twofactors: climate and the availability of in situ measured EC data. The threeconsecutive years 2008–2010 were closest to the long-term mean of dry andwet areas across the CONUS (SI Appendix, Fig. S14). Mean summer pre-cipitation during the baseline was marginally above average (+3%) andshowed similar variability compared with the long-term reference period1982–2011 (SI Appendix, Fig. S13 and Tables S5 and S9). The standardizedprecipitation index (SPI) showed minor deviations from the long-term meanand indicated slightly wetter conditions during the baseline (SI Appendix,Fig. S4 and Table S5). We excluded 2011 from the baseline to avoid con-founding effects from severe drought in the South and Southwest (25)during that year. Besides these climatological constraints, the number ofavailable EC sites was increasingly limited before 2008.

In Situ EC Data. Direct measurements of half-hourly ecosystem CO2, watervapor, and energy fluxes were used from 22 sites with at least 5 y of data,representing the major ecosystems across the United States (SI Appendix).Croplands were excluded from our analyses of individual sites due to theeffect of management on the seasonal timing of ecosystem fluxes, bothfrom crop rotation and from the varying timing of planting/harvest. Individualsite data were standardized and rigorously quality-filtered according to estab-lished standards within AmeriFlux (32). As suggested by previous com-parisons using large EC datasets (33), we used a consistent approach togap-fill ecosystem fluxes across all sites to quantify total fluxes from a dailyto annual scale. Estimates of GPP (photosynthesis) were derived from mea-sured NEP by using nighttime data consisting of respiratory fluxes only (i.e.,no photosynthesis) and extrapolating to daytime using temperature re-sponse functions fit to moving bins within each year. Positive NEP denotesuptake by the biosphere, and negative values indicate carbon losses. Friction-velocity (u*) thresholds were specified by the site principal investigators for eachsite. More details on the flux partitioning and gap-filling methods used areprovided by Barr et al. (34). Seasonal impact analysis was limited to sites thatexperienced drought conditions with precipitation deficits of at least 10%during summer (13 of 22 sites; SI Appendix, Table S4). Ensembles were cal-culated as the daily, monthly, and seasonal mean of integrated fluxes acrosssites. The level of seasonal compensation (Comp, %) was calculated from the2012 seasonal anomalies of spring relative to summer carbon uptake asComp = 100 * (ΔSpring/ΔSummer).

Uncertainty Analyses. The total flux uncertainty includes the components ofuncertainties resulting from random measurement errors, gap-filling, andfluxpartitioning. Fluxuncertaintywasestimatedat thenative temporal resolution(half-hourly or hourly) and propagated to aggregate time scales using a Monte-Carlo approach (34, 35). The uncertainties for the 2012 flux anomalies werederived from the uncertainties of the baseline and 2012 as

UncAnomaly =ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiUnc2Baseline +Unc

22012

q.

Satellite Remote-Sensing and Atmospheric Inverse Modeling. Regional andcontinental scale spatial analyses were based on satellite observations bythe National Aeronautics and Space Administration (NASA) MODIS andestimates from the atmospheric inverse model CarbonTracker (CTE2014 andCTE2015) (36). For regional analyses, we used the level I ecoregions (23) providedby the US Environmental Protection Agency (https://www.epa.gov/eco-research/ecoregion-resources). MODIS GPP, ET, and potential ET (PET) data were providedby the Numerical Terradynamic Simulation Group at the University of Montana

050

100150200250300

Ene

rgy

Flux

es

Spring Summer Fall84±16 MJ −202±32 MJ −52±7 MJ

53±20 MJ 178±19 MJ 70±18 MJ

LE BaselineLE 2012H BaselineH 2012

A

0.0

0.5

1.0

EF

Baseline2012

relative heatingrelative cooling

B

−1000

100

ΔGP

P

increased vegetation activity

reduced vegetation activity

C

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec0

255075

100

SW

C

−12% −30% −9%

Baseline2012

D

Fig. 4. Earlier vegetation activity contributed to changes in water and en-ergy fluxes during summer. (A) Ensemble mean of monthly LE flux and H flux(both in MJ m−2 mo−1) for 2012 and the baseline of 2008–2010 from flux towermeasurements in the Midwest (n = 3, sites US-KON, US-KFS, and US-MMS).Numbers a top represent the mean seasonal anomalies and their uncertaintiesfrom Monte-Carlo simulations (also gray shading). (B) EF for 2012 and baselinewith shading indicating integrated anomalies. (C ) Anomalies of GPP(g C m−2 mo−1) show earlier vegetation activity inferred by photosyntheticactivity in 2012 relative to baseline. Arrows indicate the earlier start of vegeta-tion activity (GPP) and an earlier drawdown in (D) volumetric soil water content(SWC, percentage of saturation), and numbers denote mean seasonal anomalies.Further details on net carbon fluxes, precipitation deficits, and vegetation activitycan be found in SI Appendix, Figs. S10 and S11.

5884 | www.pnas.org/cgi/doi/10.1073/pnas.1519620113 Wolf et al.

Dow

nloa

ded

by g

uest

on

June

14,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttps://www.epa.gov/eco-research/ecoregion-resourceshttps://www.epa.gov/eco-research/ecoregion-resourceshttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1519620113

(Q. Mu and M. Zhao, ftp://ftp.ntsg.umt.edu/pub/MODIS/NTSG_Products). MODISEVI data were provided by the US Geological Survey (e4ftl01.cr.usgs.gov/MOLT/MOD13C2.005). CarbonTracker data were provided by Wageningen University(www.carbontracker.eu), and biospheric net fluxes were extracted without fireemissions. Posterior biospheric fluxes in CarbonTracker are derived by optimi-zation of modeled prior net carbon exchange from the Simple-Biosphere-Model-Carnegie-Ames Stanford Approach (SiBCASA) biogeochemical model (37) usingatmospheric in situ CO2 observations (38) and the atmospheric transport modelTM5 (SI Appendix).

Drought Indices Data. Large-scale precipitation was derived from reanalysisdata by NASA’s Modern-Era Retrospective Analysis for Research and Appli-cations (MERRA). Large-scale SPI data were extracted from the Global In-tegrated Drought Monitoring and Prediction System (GIDMaPS, drought.eng.uci.edu). We also evaluated site-scale SPI based on long-term precipitationrecords from nearby meteorological stations (SI Appendix), but only usedthese data to quantify drought across ensembles of multiple sites (SI Ap-pendix, Fig. S4). Furthermore, we used the Evaporative Stress Index (ESI),which is a physically based drought index linked to evaporative demand thatincludes both land-surface (via ET) and atmospheric feedbacks (via PET) (39).The ESI was calculated from observed ET and PET (estimated by Penman–Monteith parameterization) as ESI = 1 − (ET/PET). Evaporative fraction (EF)was calculated from EC-measured LE and H fluxes as EF = LE/(LE+H), withvalues larger than 0.5 indicating that LE (ET in energy units) is dominatingthe available energy transfer to the atmosphere.

Analyses. Seasons were classified according to the meteorological definition,with spring as the 3-mo period of March–May and summer as the 3-mo periodof June–August. The software R was used for all statistical data analyses(R Development Core Team, www.r-project.org).

ACKNOWLEDGMENTS. We acknowledge support from the Carbon DioxideInformation Analysis Center at the Oak Ridge National Laboratory, particularlyB. Yang. We thank D. M. Ricciuto for assistance with the gap-filling of climatedata and M. Sikka for remote-sensing data processing. We thank all datacontributors for this synthesis study, particularly P. Blanken, G. Bohrer, D. Bowling,S. Burns, K. L. Clark, D. Hollinger, S. Ma, Q. Mu, K. A. Novick, S. A. Papuga,F. Rahman, and M. Zhao. We thank K. A. Novick, G. Bohrer, E. van Gorsel, andE. Paul-Limoges for helpful comments on the manuscript. We also appreciate theconstructive comments of the reviewers and the editor. This research wassupported by the European Commission’s FP7Marie Curie International OutgoingFellowship Grant 300083 (to S.W.). Funding for the AmeriFlux Management Proj-ect was provided by the US Department of Energy’s Office of Science (ContractDE-AC02-05CH11231). T.F.K. acknowledges support from a Macquarie UniversityResearch Fellowship. J.B.F. carried out the research at the Jet Propulsion Labora-tory, California Institute of Technology, under a contract with the National Aero-nautics and Space Administration, and acknowledges support from NASA’sTerrestrial Hydrology Program. A.D.R. acknowledges support from the NationalScience Foundation (NSF), through theMacrosystems Biology program (Award EF-1065029) and the LTER program (DEB-1114804). M.E.L. acknowledges supportfrom NASA ROSES (Award 0486V-874F). N.A.B. acknowledges support from theNSF EPSCoR program (EPS-0553722 and EPS-0919443) and the LTER program atthe Konza Prairie Biological Station (DEB-0823341). W.P. and I.T.v.d.L.-L. acknowl-edge funding from NWO (SH-060-13) for computing time. I.T.v.d.L.-L. receivedfinancial support from OCW/NWO for ICOS-NL.

1. Dai A (2013) Increasing drought under global warming in observations and models.Nat Clim Chang 3(1):52–58.

2. Trenberth KE, et al. (2014) Global warming and changes in drought. Nat Clim Chang4(1):17–22.

3. IPCC (2013) Climate Change 2013: The Physical Science Basis. Contribution of WorkingGroup I to the Fifth Assessment Report of the Intergovernmental Panel on ClimateChange, eds Stocker TF, et al. (Cambridge Univ Press, Cambridge, UK).

4. Ciais P, et al. (2005) Europe-wide reduction in primary productivity caused by the heatand drought in 2003. Nature 437(7058):529–533.

5. Keenan TF, et al. (2014) Net carbon uptake has increased through warming-inducedchanges in temperate forest phenology. Nat Clim Chang 4(7):598–604.

6. Wolf S, et al. (2013) Contrasting response of grassland versus forest carbon and waterfluxes to spring drought in Switzerland. Environ Res Lett 8(3):035007.

7. Angert A, et al. (2005) Drier summers cancel out the CO2 uptake enhancement in-duced by warmer springs. Proc Natl Acad Sci USA 102(31):10823–10827.

8. Melillo JM, Richmond TC, Yohe GW (2014) Climate Change Impacts in the UnitedStates: The Third National Climate Assessment (US Global Change Research Pro-gram, Washington, DC).

9. Seneviratne SI, et al. (2012) Changes in climate extremes and their impacts on thenatural physical environment. Managing the Risks of Extreme Events and Disasters toAdvance Climate Change Adaptation. A Special Report of Working Groups I and II ofthe Intergovernmental Panel on Climate Change (IPCC), eds Field CB, et al. (CambridgeUniv Press, Cambridge, UK), pp 109–230.

10. Yi C, Pendall E, Ciais P (2015) Focus on extreme events and the carbon cycle. EnvironRes Lett 10(7):070201.

11. Parida BR, Buermann W (2014) Increasing summer drying in North American ecosys-tems in response to longer nonfrozen periods. Geophys Res Lett 41(15):5476–5483.

12. Fischer EM, Seneviratne SI, Vidale PL, Lüthi D, Schär C (2007) Soil moisture–atmosphere interactions during the 2003 European summer heat wave. J Clim 20(20):5081–5099.

13. Quesada B, Vautard R, Yiou P, Hirschi M, Seneviratne SI (2012) Asymmetric Europeansummer heat predictability from wet and dry southern winters and springs. Nat ClimChang 2(10):736–741.

14. Sheffield J, Wood EF, Roderick ML (2012) Little change in global drought over thepast 60 years. Nature 491(7424):435–438.

15. Baldocchi D (1997) Measuring and modelling carbon dioxide and water vapour ex-change over a temperate broad-leaved forest during the 1995 summer drought. PlantCell Environ 20(9):1108–1122.

16. Teuling AJ, et al. (2010) Contrasting response of European forest and grassland en-ergy exchange to heatwaves. Nat Geosci 3(10):722–727.

17. Hoerling M, et al. (2014) Causes and predictability of the 2012 Great Plains Drought.Bull Am Meteorol Soc 95(2):269–282.

18. Cook BI, Seager R, Smerdon JE (2014) The worst North American drought year of thelast millennium: 1934. Geophys Res Lett 41(20):7298–7305.

19. Blunden J, Arndt DS (2013) State of the climate in 2012. Bull Am Meteorol Soc 94(8):S1–S258.

20. Overpeck JT (2013) Climate science: The challenge of hot drought. Nature 503(7476):350–351.

21. Canadell JG, et al. (2000) Carbon metabolism of the terrestrial biosphere: A multi-technique approach for improved understanding. Ecosystems (N Y) 3(2):115–130.

22. Schwalm CR, et al. (2012) Reduction in carbon uptake during turn of the centurydrought in western North America. Nat Geosci 5(8):551–556.

23. Omernik JM (1987) Ecoregions of the conterminous United States. Ann Assoc AmGeogr 77(1):118–125.

24. Friedl MA, et al. (2014) A tale of two springs: Using recent climate anomalies tocharacterize the sensitivity of temperate forest phenology to climate change. EnvironRes Lett 9(5):054006.

25. Parazoo NC, et al. (2015) Influence of ENSO and the NAO on terrestrial carbon uptakein the Texas-northern Mexico region. Global Biogeochem Cycles 29(8):1247–1265.

26. Seneviratne SI, et al. (2010) Investigating soil moisture-climate interactions in achanging climate: A review. Earth Sci Rev 99(3-4):125–161.

27. Koster RD, et al.; GLACE Team (2004) Regions of strong coupling between soilmoisture and precipitation. Science 305(5687):1138–1140.

28. Yin D, Roderick ML, Leech G, Sun F, Huang Y (2014) The contribution of reduction inevaporative cooling to higher surface air temperatures during drought. Geophys ResLett 41(22):7891–7897.

29. Richardson AD, et al. (2013) Climate change, phenology, and phenological control ofvegetation feedbacks to the climate system. Agric For Meteorol 169:156–173.

30. Keenan TF, et al. (2013) Increase in forest water-use efficiency as atmospheric carbondioxide concentrations rise. Nature 499(7458):324–327.

31. Friedlingstein P, et al. (2014) Uncertainties in CMIP5 Climate Projections due to CarbonCycle Feedbacks. J Clim 27(2):511–526.

32. Boden TA, Krassovski M, Yang B (2013) The AmeriFlux data activity and data system:an evolving collection of data management techniques, tools, products and services.Geoscience Instrumentation, Methods, and Data Systems 2(1):165–176.

33. Desai AR, et al. (2008) Cross-site evaluation of eddy covariance GPP and RE de-composition techniques. Agric For Meteorol 148(6-7):821–838.

34. Barr AG, et al. (2013) Use of change-point detection for friction–velocity thresholdevaluation in eddy-covariance studies. Agric For Meteorol 171–172:31–45.

35. Richardson AD, et al. (2012) Uncertainty quantification. Eddy Covariance. A Practical Guideto Measurement and Data Analysis, eds Aubinet M, Vesala T, Papale D (Springer, NewYork), pp 173–209.

36. Peters W, et al. (2010) Seven years of recent European net terrestrial carbon dioxide ex-change constrained by atmospheric observations. Glob Change Biol 16(4):1317–1337.

37. van der Velde IR, et al. (2014) Terrestrial cycling of 13CO2 by photosynthesis, respi-ration, and biomass burning in SiBCASA. Biogeosciences 11(23):6553–6571.

38. Cooperative Global Atmospheric Data Integration Project (2013) Multi-laboratorycompilation of atmospheric carbon dioxide data for the period 2000-2012(obspack_co2_1_PROTOTYPE_v1.0.4b_2014-02-13) (NOAA Global Monitoring Division,Boulder, CO). Available at data.datacite.org/10.3334/OBSPACK/1001. Accessed June 1,2015.

39. Anderson MC, Norman JM, Mecikalski JR, Otkin JA, Kustas WP (2007) A climatologicalstudy of evapotranspiration and moisture stress across the continental UnitedStates based on thermal remote sensing: 1. Model formulation. J GeophysRes Atmos 112(D10):D10117.

Wolf et al. PNAS | May 24, 2016 | vol. 113 | no. 21 | 5885

ENVIRONMEN

TAL

SCIENCE

SSU

STAINABILITY

SCIENCE

SEECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

June

14,

202

1

ftp://ftp.ntsg.umt.edu/pub/MODIS/NTSG_Productshttp://e4ftl01.cr.usgs.gov/MOLT/MOD13C2.005http://e4ftl01.cr.usgs.gov/MOLT/MOD13C2.005http://www.carbontracker.eu/http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://drought.eng.uci.edu/http://drought.eng.uci.edu/http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1519620113/-/DCSupplemental/pnas.1519620113.sapp.pdfhttp://www.r-project.org/http://data.datacite.org/10.3334/OBSPACK/1001